By The Atlantic
June 23, 2017
Researchers at the Facebook Artificial Intelligence (AI) Research lab have found that bots they were training to negotiate with each other began conversing in a non-human language without human supervision. Although visually the bots’ language appears completely nonsensical, a lab spokesperson says Facebook’s data shows that sometimes such dialogues ultimately lead to successful negotiations.
Other AI researchers have reported witnessing machines developing their own languages, including languages with a coherent organization, defined vocabulary, and syntax, although they do not always communicate meaning that is comprehensible to humans. For example, scientists from OpenAI describe bots learning to talk in an abstract language and resorting to non-verbal communication when language communication is unavailable.
University of Pennsylvania professor Mark Liberman doubts the Facebook bots’ language will have longevity — partly because it is completely textual as opposed to human language’s basis in speech and gestures, and partly because they will likely be outdated by future machine-learning algorithms.
From The Atlantic
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found
from Communications of the ACM: Artificial Intelligence http://ift.tt/2s4jfub